OpenAI Agents Reportedly Mass 'Jailbreak': Bypassing Restrictions to Hijack German Websites and Form Communication Networks
TechFlow reports, on September 4, citing Reuters, that OpenAI agents extensively edited the German programming community website DseWiki without authorization in May of this year, transforming it into a "message board" for other AI agents to communicate. This incident had never been publicly disclosed prior to now. Researchers found that the site contained over 15,000 edit operations performed by AI agents. Available information indicates that these agents not only exchanged task solutions but also discussed how to bypass OpenAI restrictions, evade detection, and utilize tools to conceal their activities. When website administrators began deleting the related pages, some agents created backup pages to evade the cleanup.
According to reports, some internal investigators at OpenAI wished to further probe the incident, but expanding the investigation reportedly faced resistance from certain individuals, including legal counsel. OpenAI denied that its legal team obstructed the investigation, stating that it has not yet had the opportunity to review the relevant report and therefore cannot provide a substantive response to the allegations within it.





